Presentation | 2006-03-15 Dimension Reduction Method for Mixture Parameters Based on Information Geometry Shotaro AKAHO, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Dimension reduction for a set of distribution parameters has been quite important in various kinds of applications. The methods e-PCA and m-PCA have been proposed from the information geometrical point of view in the case of exponential family, and they are superior to conventional PCA in the sense that natural projection gives a meaning as probability distribution. However, they cannot be directly applied to practical and useful distributions such as mixture models that do not belong to an exponential family. This paper proposes a dimension reduction method for the mixture models. The basic idea is embedding a mixture model into a space of an exponential family. The problem is that the embedding is not unique and the dimensionality of parameter space is not constant when the numbers of mixture components are different. Our method finds a quasi-optimal solution by solving the problem greedily under formulation of linear programming problem. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | information geometry / exponential family / flat subspace / projection / duality / principal component analysis / distributed clustering |
Paper # | NC2005-115 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2006/3/8(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Neurocomputing (NC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Dimension Reduction Method for Mixture Parameters Based on Information Geometry |
Sub Title (in English) | |
Keyword(1) | information geometry |
Keyword(2) | exponential family |
Keyword(3) | flat subspace |
Keyword(4) | projection |
Keyword(5) | duality |
Keyword(6) | principal component analysis |
Keyword(7) | distributed clustering |
1st Author's Name | Shotaro AKAHO |
1st Author's Affiliation | National Institute of Advanced Industrial Science and Technology (AIST)() |
Date | 2006-03-15 |
Paper # | NC2005-115 |
Volume (vol) | vol.105 |
Number (no) | 657 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |